Webhooks made sense when the consumer was a cloud server with a static IP and a permanent HTTPS endpoint. For AI agents, almost none of those assumptions hold. Most agents run behind NAT. Laptops behind home routers. Docker containers behind corporate firewalls. Cloud VMs with no public IP (which is now the default on most cloud providers for sensible security reasons). None of these can receive webhooks without building additional infrastructure first. I spent a while doing it the hard way. Here is what changed when I stopped. The actual cost of doing webhooks properly The frustrating thing about webhooks is that the naive implementation looks simple: accept a POST, process the payload, return 200. The production implementation is a different story.…